Akash notebooks

[Shrike05]

TL;DR

Akash Notebooks (AKN) is an online platform for data science and machine learning where users can train models, compete with models, share datasets and collaborate on projects.
Think Kaggle or Google Colab, but powered by Akash Network. The solution is meant to abstract the complexity, enabling advanced GPU-accelerated development where you pay-as-you-go. In addition, the competitions feature primes Akash Notebooks to stand out as the platform for AI and DeAI community’s to create and manage competitions.

Moreover Akash Notebooks serves as a community tool for current Insiders programs such as Ringmasters for creating and managing events such as competitions, for potential Akash University Program and for anyone interested in promoting Akash Network.

Motivation

With Akash Network's decentralized compute infrastructure the sky is the limit. Even so the onboarding experience can still feel technically intense, with pre-requisites like dockerization, deployment management, and networking skills. This often impedes data scientists, students, and researchers who lack the DevOps knowledge. Current mainstream alternatives solve this, but offer limited session time, no file system, and inferior hardware with binding subscriptions. This stifles innovation and experimentation beyond simple models.
Akash Notebooks addresses this gap by offering a user interface for long-form model training, code access, and a community-driven competitions ecosystem. The goal is to significantly lower the barrier to entry for leveraging Akash's GPU compute and attract a wider audience to Akash Network.

Technical Description

Akash Notebooks consists of the following components:

• Dockerized Jupyter Lab Instance: A forked and branded version of Jupyter Lab customized for Akash Notebooks, packaged in a container with GPU access.
• Integration with GitHub: Allowing users to automatically load their GitHub code into Jupyter Lab as it is deployed to get started immediately.
• Deployment Dashboard GUI: A web-based frontend that abstracts Akash SDL and deployment management, enabling users to launch and manage compute jobs without CLI or YAML configurations.
• Filesystem Integration: The platform integrates with users' GitHub repositories and exposes them through a drag-and-drop folder UI mimicking a traditional OS environment.
• VS Code Support: Enables opening Jupyter Lab instances within the user's local VS Code setup for a familiar development experience.
• Persistent Session Management: Users fund deployments directly, allowing them to run long training sessions uninterrupted as opposed to time-limited sessions on competing platforms.
• Event Management Module: A database and API backend powers an event section.
• Dataset and Pipeline Service: Users can create a pipeline that reads a dataset, processes or transforms it, and outputs a new version of the dataset. This enables reproducible and shareable data transformation workflows.

Current Status

A minimal viable product (MVP) of Akash Notebooks has been built with 2-3 core features which will be presented through a demo video one week from posting this discussion. The open-source license for the project is still to be determined. Once a future proposal is approved and application is complete, we will open-source the GitHub repository.
Authentication & Payments
Users can authenticate using either their Keplr wallet, GitHub OAuth or Email. Users may choose to fund their deployments in either AKT or USD, accommodating both crypto-native and traditional users.

Target Audience

The initial focus is on individuals such as students, researchers, data scientists, and independent ML engineers. However, the platform’s aim is to pivot toward team and organisational use cases.

Benefits to Akash Network

• Increased Akash Network Utilization: Encourages broader use of Akash Network resources by making them more accessible.
• Expanded User Base: Opens Akash Network to non-developers such as data scientists, researchers, students, and startups.
• Improved Brand Visibility: Acts as a community tool for Ringmasters and potential Akash Insiders University Program to run workshops or competitions.
• Provider Incentives: Increases opportunity for providers to monetize Akash Network.
• Support for Open Source: Reinforces Akash Network’s commitment to open and decentralized systems.

Open Questions / Request for Feedback

• What would be the best way to encourage community contributions to this open-source project?
• How to best monetize the app to make it self-sustained?
• How soon should we strive to integrate FIAT?

Demo Video:
https://www.youtube.com/watch?v=qryk7qxpDuo

HardBoiledEggs
Akash Insider
M10 AI

[oiclid]

This is great and timely. It would hopefully provide open-source and affordable alternatives to Google Colab, Kaggle, and Paperspace Gradient.

To answer your questions:

1. Clear contributor guidelines, starter issues, bounties, leaderboards, hackathons, and grants should help pique interest.
2. Using a freemium model and/or subscription-based model seems like a great way to start.
3. I think waiting until the system is relatively mature (3-6 months of use) would be a good time to start working on Fiat integrations.

Good job!

[Shrike05]

Hey

We have been thinking of creating a good library of issues that people can easily pick up on and work. As for hackathons that is very much the plan, similar to kaggle we want people to be able to write code and directly submit the work to the hackathon. Bounties, leaderboards and grants are also things to think about. Thank you for the suggestion.

Freemiums were also a part of the plan to try to get a first user base on board, how we will go about with freemiums is something we are still discussing however because we obviously don't want people taking advantage of the system.

We want to introduce fiat relatively soon but it definitely makes sense that the system should at least be mature before we start adding auxiliary features like that.

Thanks for your input :)

[rianfloo]

Hey! This is a super cool and ambitious project! 😄

Just to answer to your questions, I think it’d help to clarify upfront whether it’s open source or not, before diving into how people can contribute. If it is open, maybe think about a way contributors could earn credits or some kind of perk.

Since your target audience is mostly researchers, students, etc., chances are they won’t have a ton of budget. Even if your product is awesome, they’ll probably lean toward the most affordable or free option. So having a killer feature, something others aren’t offering could really set you apart.

As for adding FIAT payments, it’s not a must-have for the MVP, but I’d definitely consider it down the line. Sticking only to crypto could limit your user base and make it harder to grow.

Happy to help you!

Good luck!

[Shrike05]

Hey thanks for the feedback

We are planning to make it open source in the future if we get the proposal approved and we can maintain it. About students and researchers not having the budget, we are actually planning to include freemiums for that exact reason and we believe it will be worth it to continue using the platform because we do have a number of features that set us apart from the likes of colab and kaggle like persistent storage, a file system and better spread of gpus (like H200). Not to mention that this becomes cheaper in the end because instead of paying for subscriptions you pay only for what you use.

[Fluffy9]

Hey, I've been wanting to have community crowd code events for a while and have been looking for a good project that the core team isn't working on. I could see this one being a good candidate. If the community is supportive of this, I would love to have this be something we work on bi-weekly with contributions program funding

[Shrike05]

Im happy to see the enthusiasm. We will work hard to make this app as best as it can possibly be

[piber-dev]

hello @Shrike05 thanks for writing this! i really like the idea of this, i hope to see the MVP soon.

What would be the best way to encourage community contributions to this open-source project?

by open-sourcing it, writing clear code, and writing contribution docs! i'm not sure how easy it will be to get a lot of community contributions to the platform but if you are able to get a separate bounty budget specifically for this or work with the community contributions program or zealy program you might get a few extra pair of eyes interested.

How to best monetize the app to make it self-sustained?

the more freemium you go, the riskier it becomes. the most predictable way is to only have paid tiers or requesting funds from the community pool that covers the expenses for the freemium users. whether or not the rest of the community would like to fund these features is impossible to say though. hopefully you can create a business plan that you can be proud over with your philosophy without it bankrupting your wallets.

How soon should we strive to integrate FIAT?

of course you need to make design choices to make this integration as simple as possible but i would not focus on this until after the platform is complete. receiving money adds complexity and tax burdens, which i would not want to care about when the userbase is low or non-existent.

[Shrike05]

Thanks piber-dev this was very useful feedback
We are ofcourse planning to make it easier for people to contribute to the codebase through clear guidelines but working together with community contributions or zealy is a very interesting idea aswell. As for freemiums, they are meant to be a way to get people to try the application first and spread it within their own groups, we will likely keep them quite exclusive and sparse to avoid running out of money. Lastly fiat integration would definietly add a lot more complexity to the product so we will focus on functionality first, you also mentioned tax burdens which I 100% agree with you there, but we still think it is a very necessary features to have if we want to go mainstream since most students, researchers and the like aren't crypto-friendly. Thank you for the support.

[production23]

Hey HardBoiledEggs 👋,

This looks like a strong step forward for DeAI infrastructure — excited to see Akash Notebooks taking shape!

I'm part of a small team focused on data labeling and annotation tooling (video/image/LLM/3D), and we’re currently offering free pilot services to open-source or early-stage AI platforms like this. If you're planning any competitions or demos involving model training, we’d be happy to help with:

🏷️ High-quality dataset labeling (COCO, VOC, JSON, etc.)

✅ Review pipelines with QA and gold standard validation

🎯 Multi-format export and dashboard analytics

We’re also experimenting with integrations via API and could potentially sync with your GitHub-based pipeline or notebook infrastructure.

Let me know if this could be useful during your rollout or demo week — happy to send a sample, contribute, or connect directly.

Best,
Evolearns Team [email protected]

[Shrike05]

Hey Evolearns thank you for reaching out. We are still in the early stages of development but once we have matured consistent hackathons and competitions are part of the plan to create a strong community. We would be delighted for a partnership when that time comes and an API integration to simplify using the data sounds like a great idea

[Shrike05]

The demo video will unfortunately be late because of unforeseen errors. We will get back with the demo as soon as possible

[harystyleseze]

I’m really excited about the direction of Akash Notebooks, especially the features around dataset sharing and project collaboration, which are essential for building a strong ML and AI community. I have a few ideas that might further enhance the platform: First, supporting the sharing of reusable ML stack configurations—like container setups or YAML templates—would allow users to quickly deploy common environments without needing deep DevOps skills, more like a marketplace. Second, what do you think about enabling users to publish training logs or results on-chain? This could boost transparency, verifiability, and support decentralized AI efforts, but I am also looking at the complexity. Lastly, regarding GPU access, a community-driven GPU compute sharing portal (a dedicated portal within the platform), where individuals can rent out or contribute their idle GPU resources, might expand compute availability and create new incentives for participation. These ideas could significantly strengthen Akash Notebooks as a truly decentralized and collaborative AI platform.

[Shrike05]

Hey thanks for the suggestions, you have some very unique ideas that are definitely worth considering. being able to share stack configurations strikes me as one that would be quite simple to implement yet very powerful. As for enabling users to publish training logs and results on-chain I have difficulty imagining how that would look but I would love to get in touch so I can hear what you're thinking. I'm not sure how a GPU portal might look because Akash needs a certain level of prerequisite setup before being able to provide compute on the network so a portal through the website might be nigh impossible unless we setup our own system separate from Akash.
Thanks for the suggestions I would love to hear more from you again

[baktun14]

This is cool and thanks for your contribution to the Akash ecosystem! However, I'm re-iterating the fact that I don't see that value in replicating the user experience already available in Akash Console to create and manage deployments. It should be an extra feature specific of notebooks directly in Console. It would also benefit for credit card users in Console.

[Shrike05]

This is actually a major point I wanted to bring fourth, that this isn't just a akash console copy. Ofcourse the main objective requires that we emulate a lot of akash console features but, the ultimate goal of akash notebooks is to do whatever else is necessary for a datascientist looking to use the service. This means that it should be simpler to mount data on the jupyter client, it should be simple to continue working on a project by saving it to github and automatically loading it into the notebook on startup (note I am not talking about 'deploying' the github repository). We are also planning features to make it simpler for teams to collaborate on projects and for ai devs to use federated training. The main purpose will ofcourse still be "Manage deployments of jupyter instances" which would make it seem like a worse Akash console, but the value of the app lies in improving the developer experience with all the extra nonsense data scientists would have to deal with outside of that main purpose.

[koop46]

Akash Console is not a ML development environment and community hub it. It's a deployment management tool and can therefore coexist.

[baktun14]

I definitely see the value into bringing vertical use cases for Akash. Akash Console is like the AWS/GCP/Azure frontend to use Akash and it means that it can have vertical products within the platform. The idea is to also not duplicate and spread open source work related to deploying on Akash, but rather merge into the ultimate deployment experience. Perhaps this project can serve as a pilot program to experiment this kind of vertical, see the interest and then if sticks we can integrate it in console.

I've seen too many community pool proposals just come and go, either unmaintained, abandonned or unfinished so I'm always critical with new ideas.

[anilmurty]

Hey there!

Thanks for the thoughtful proposal and demo! - pretty cool and somewhat similar to https://github.com/orgs/akash-network/discussions/609

My comments/ questions:

Product:

• If the goal is to appeal to researchers and student (which i believe is the primary user persona for google collab) then I think Wallet/ Crypto based product is a no-go and if you have to get them to use this instead of google colab (which is free) then there needs to be a reason for them to do that. What is the that reason?

• Further, the demo seems to implement a new UI around the Jupyter notebook templates that exist in Console/ Awesome-akash (eg https://console.akash.network/templates/akash-network-awesome-akash-jupyter) -- what is the need to create a new UI? Aren't you better off just adding some new templates in Akash that include various libraries (pytorch, TF etc)? Console also supports deployments via a credit card (or crypto) so there's that benefit also.

• As for solving the need for dockerization, there has already been a feature build that takes a react code base and dockerizes and deploys it without needing to know how to use docker or write YAML. I would think a similar approach to that that lives as a template in Console may be the right way to go here? have you tried using: https://console.akash.network/new-deployment?step=edit-deployment&templateId=akash-network-awesome-akash-automatic-deployment-CICD-template&gitProvider=github ?
  Have you considered building something like that but for Jupyter notebooks?
  Separately, the core team is also working with a team (levangie labs who presented at Accelerate recently) who is prototyping a AI agent based deployment workflow that is able to build a docker container and SDL with natural language inout.

GTM:
Google Collab is primarily targeted at researchers and students. Couple things to think about there: 1) What is the big value add this offers over google colab (which is feature rich and free)? and 2) Assuming that it does, how would we take this solution to them?